Fault Detection Method Based on Siamese Convolutional Neural Networks

Bearing components are important parts of mechanical equipment and have a significant impact on the operating status of the equipment. However, in actual industrial production, the data generated by bearing components is characterized by high dimensionality, nonlinearity, insufficient fault data, noise content, and operating load variation. Traditional machine learning methods for fault detection require human selection of fault features, and their reliability cannot be guaranteed. Therefore, to address the above problems, this paper proposes a fault detection model based on Siamese Convolutional Neural Networks (Siamese CNN) and conducts experiments on the benchmark bearing fault dataset collected by the Case Western Reserve University (CWRU) data center. The results show that the average detection accuracy of the model reaches 99.6% under four working conditions, which is 0.4% higher than that of the Convolutional Neural Networks (CNN) model; the detection accuracy of the model is still higher than that of the CNN model in the case of fewer training samples and stronger noise; the average detection accuracy of the Siamese CNN model reaches 98.87%, which is 3.98% higher than that of the CNN model under six variable load cases and the detection accuracy of the model reaches 97% for all of the six variable load The detection accuracy of the Siamese CNN model reaches 98.87% on average, which is 3.98 % higher than that of the CNN model, and the detection accuracy in the six variable load cases reaches over 97%. The experiment fully verifies the effectiveness of the model.